Scoring & lanes

Every audit runs the same 26 deterministic checks against three fetches: the page you submit, the origin's robots.txt, and the origin's llms.txt. No LLM calls, no sampling — the same page state always scores the same.

How is the score computed?

Each lane's score is simply the percentage of its checks that pass. The overall score is the weighted average of the five lanes:

Lane	Weight	Checks
AI Crawler Access	25	7
Citability	25	5
Metadata & Sharing	20	5
Structured Data	20	3
Trust Signals	10	3

Worked example: if you score 100 on everything except Structured Data at 33 (one of three checks passing), the overall is (25·100 + 25·100 + 20·100 + 20·33 + 10·100) / 100 ≈ 87 → grade B.

What are the grade bands?

• A — 90 to 100
• B — 80 to 89
• C — 65 to 79
• D — 50 to 64
• F — below 50

Lane 1 — AI Crawler Access (weight 25)

Whether AI engines are allowed to read you at all. If robots.txt is missing entirely, all crawlers are allowed by default and the lane treats that as open access.

• GPTBot allowed — OpenAI's crawler (training and, increasingly, retrieval for answers).
• ClaudeBot allowed — Anthropic's crawler.
• Claude-Web allowed — Anthropic's on-demand fetcher.
• PerplexityBot allowed — Perplexity's index crawler.
• Google-Extended allowed — controls Gemini training use of your content.
• CCBot allowed — Common Crawl, an upstream corpus for many models.
• llms.txt present — a curated markdown map of your site for AI consumers, served at /llms.txt. The check requires a non-empty file that isn't an HTML error page.

A bot counts as blocked when a User-agent section naming it contains Disallow: / (a later Allow: / in the same section unblocks it). Note the parser matches sections that name the bot explicitly — a blanket User-agent: * block is a broader problem an engine-specific check doesn't flag.

Lane 2 — Metadata & Sharing (weight 20)

What engines and link previews see before reading a word of your content.

• Title tag — present and 15–70 characters.
• Meta description — present and 50–160 characters. Answer engines frequently lift these verbatim.
• Canonical URL — a <link rel="canonical"> so citations consolidate to one URL.
• Open Graph tags — both og:title and og:image present.
• HTTPS — the final URL after redirects resolves over HTTPS.

Lane 3 — Structured Data (weight 20)

JSON-LD is how engines verify who you are rather than guessing from prose.

• JSON-LD structured data — at least one valid application/ld+json block declaring a type.
• Entity schema — an Organization, Person, or LocalBusiness node.
• FAQ schema — a FAQPage node; question-answer pairs are the most directly extractable format for AI answers.

Lane 4 — Citability (weight 25)

Whether there's a clean passage for an engine to quote.

• Single clear H1 — exactly one <h1> on the page.
• Section structure — at least two <h2> headings; passages need boundaries.
• Content depth — at least 300 words of visible text after stripping markup.
• Lists or tables — at least one <ul>, <ol>, or table-like structure.
• Question-form headings — at least one H2/H3 phrased as a question (contains "?", or starts with how/what/why/when/which).

Lane 5 — Trust Signals (weight 10)

Cheap-to-fake signals are weighted low, but their absence still costs citations.

• About page linked — a link whose href contains /about.
• Contact method visible — a /contact link or a mailto: address.
• Dates / freshness signals — <time> elements, datetime attributes, or datePublished/dateModified markup.

How are the "fix these first" items ranked?

Every failed check carries a suggested fix. Fixes are sorted by their lane's weight (heaviest lanes first) and the top five become your priority list. Fixing in that order moves the score fastest.

Limits worth knowing

• The audit inspects the URL you submit, not your whole site. Run it on your homepage and your key landing pages separately.
• Fetches follow up to 5 redirects, time out at 10 seconds per resource, and cap responses at 4 MB.
• Server-side rendering matters: checks read the HTML your server returns. Content that only appears after client-side JavaScript runs is invisible to the auditor — and to most crawlers.